incorrectly labels normal epithelial cells as tumor cells

Dear developers,

I'm testing ikarus using data (KUL3-21N and KUL3-21T) from "Lee, Hae-Ock, et al. "Lineage-dependent gene expression programs influence the immune landscape of colorectal cancer." Nature Genetics 52.6 (2020): 594-603."

I noticed that when I used normal samples, ikarus still labels normal epithelial cells as tumor cells, any suggestions for that?

The analysis can be found in https://github.com/neil-n-zhang/ikarus_LeeCRC/tree/main/code

Thanks so much! Neil

Hi Neil Does it label all of them as tumor or some of them?

On Thu, Jul 7, 2022 at 12:05 AM Neil Zhang @.***> wrote:

Dear developers,

I'm testing ikarus using data from "Lee, Hae-Ock, et al. "Lineage-dependent gene expression programs influence the immune landscape of colorectal cancer." Nature Genetics 52.6 (2020): 594-603."

I noticed that when I used normal samples, ikarus still labels normal epithelial cells as tumor cells, any suggestions for that?

The analysis can be found in https://github.com/neil-n-zhang/ikarus_LeeCRC/tree/main/code

Thanks so much! Neil

— Reply to this email directly, view it on GitHub https://github.com/BIMSBbioinfo/ikarus/issues/13, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAE32EKKENCYIZ4QURPNGTDVSX7MHANCNFSM523JGZXQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Sent from mobile, excuse the brevity

Hey Neil,

Thank you so much for taking interest in ikarus!

Lee dataset is a bit weird in a sense that we never managed to get exceptionally good testing/validation performance on the data. We are not certain whether the labels are imprecise, whether there was an additional technical artifact that increased the similarity of the cancer/epithelial transcriptomes, or whether it's a peculiarity of the data/cancer type that there is a strict transcriptomic/genetic continuum between the epithelial and transformed cells.

We even tried increasing the stringency of labels by looking only at cells which had uniformly high cancer hallmark scores - we are in the process of adding both the scripts and the labels to the repository.

However, when combined with other datasets for extraction of invariant features, and definition of the threshold boundary of the logistic model, the Lee dataset showed amazing performance - which means that the dataset really contains all the relevant information, it's just fuzzy.

Please let me know if this answers your question.

Best,

Vedran

On Thu, Jul 7, 2022 at 9:24 AM Altuna Akalin @.***> wrote:

Hi Neil Does it label all of them as tumor or some of them?

On Thu, Jul 7, 2022 at 12:05 AM Neil Zhang @.***> wrote:

Dear developers,

I'm testing ikarus using data from "Lee, Hae-Ock, et al. "Lineage-dependent gene expression programs influence the immune landscape of colorectal cancer." Nature Genetics 52.6 (2020): 594-603."

I noticed that when I used normal samples, ikarus still labels normal epithelial cells as tumor cells, any suggestions for that?

The analysis can be found in https://github.com/neil-n-zhang/ikarus_LeeCRC/tree/main/code

Thanks so much! Neil

— Reply to this email directly, view it on GitHub https://github.com/BIMSBbioinfo/ikarus/issues/13, or unsubscribe < https://github.com/notifications/unsubscribe-auth/AAE32EKKENCYIZ4QURPNGTDVSX7MHANCNFSM523JGZXQ

. You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Sent from mobile, excuse the brevity

— Reply to this email directly, view it on GitHub https://github.com/BIMSBbioinfo/ikarus/issues/13#issuecomment-1177188616, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAGY7CBIRNAXBWSG4I4IBETVS2A25ANCNFSM523JGZXQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Vedran Franke Bioinformatics and Omics Data Science Platform Berlin Institute for Medical Systems Biology (BIMSB) Max Delbrück Center (MDC) Hannoversche Straße 28

Web:http://bioinformatics.mdc-berlin.de

Hi @al2na , it labels all normal epithelial cells as tumor, the non-epithelial cells as normal in the normal sample.

Hi @frenkiboy, thanks so much for your reply and looking forward to the updates. Two more suggestion: 1. From the paper, I saw you tested ikarus using healthy PBMC samples. It will be better to also add the specificity test using normal epithelial cells. It will help us make sure we have a tumor cells classifier rather than general epithelial cells classifier. 2. Even for the epithelial cells from the tumor, there are some normal epithelial cells. It may be better to only use the cells with high cancer hallmark scores as tumor cells during training.

Thanks so much! Neil

Hey Neil,

I completely agree with you! We actually planned to test it on normal epithelial cells, however, we ended up testing it on a visium breast cancer sample provided by 10x, and it performed quite well. This part didn't end up in the paper, but should be in the response to the reviewers. Unfortunately, by the end , we were lacking manpower to push additional analysis. In general, the intuition we have now is that indirectly we are indirectly measuring genetic changes - regions that are common copy number aberrations across cancers, which would imply that the method should discriminate normal epithelial from cancer cells (there is always the possibility that some of the epithelial lineages have genetic abnormalities). In a recent talk Christina Curtis showed quite convincing data that gastric cancer epithelial cells evolve through deterministic genetic changes (chromosome rearrangements/gain/loss), which makes me a bit more confident that pan cancer type genetic aberrations might not be so uncommon (of course, constrained by the developmental lineage).

On Thu, Jul 7, 2022 at 7:34 PM Neil Zhang @.***> wrote:

Hi @al2na https://github.com/al2na , it labels all normal epithelial cells as tumor, the non-epithelial cells as normal.

Hi @frenkiboy https://github.com/frenkiboy, thanks so much for your reply and looking forward to the updates. Two more suggestion: 1. From the paper, I saw you tested ikarus using healthy PBMC samples. It will be better to also add the specificity test using normal epithelial cells. It will help us make sure we have a tumor cells classifier rather than general epithelial cells classifier. 2. Even for the epithelial cells from the tumor, there are some normal epithelial cells. It may be better to only use the cells with high cancer hallmark scores as tumor cells during training.

Thanks so much! Neil

— Reply to this email directly, view it on GitHub https://github.com/BIMSBbioinfo/ikarus/issues/13#issuecomment-1177976861, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAGY7CFHV3EIXRIHJP6EIUDVS4IJ7ANCNFSM523JGZXQ . You are receiving this because you were mentioned.Message ID: @.***>

Thanks so much for the insights, I will close the issue for now. Looking forward to more ikarus updates in the future!

Hey Neil, I completely agree with you! We actually planned to test it on normal epithelial cells, however, we ended up testing it on a visium breast cancer sample provided by 10x, and it performed quite well. This part didn't end up in the paper, but should be in the response to the reviewers. Unfortunately, by the end , we were lacking manpower to push additional analysis. In general, the intuition we have now is that indirectly we are indirectly measuring genetic changes - regions that are common copy number aberrations across cancers, which would imply that the method should discriminate normal epithelial from cancer cells (there is always the possibility that some of the epithelial lineages have genetic abnormalities). In a recent talk Christina Curtis showed quite convincing data that gastric cancer epithelial cells evolve through deterministic genetic changes (chromosome rearrangements/gain/loss), which makes me a bit more confident that pan cancer type genetic aberrations might not be so uncommon (of course, constrained by the developmental lineage). … On Thu, Jul 7, 2022 at 7:34 PM Neil Zhang @.> wrote: Hi @al2na https://github.com/al2na , it labels all normal epithelial cells as tumor, the non-epithelial cells as normal. Hi @frenkiboy https://github.com/frenkiboy, thanks so much for your reply and looking forward to the updates. Two more suggestion: 1. From the paper, I saw you tested ikarus using healthy PBMC samples. It will be better to also add the specificity test using normal epithelial cells. It will help us make sure we have a tumor cells classifier rather than general epithelial cells classifier. 2. Even for the epithelial cells from the tumor, there are some normal epithelial cells. It may be better to only use the cells with high cancer hallmark scores as tumor cells during training. Thanks so much! Neil — Reply to this email directly, view it on GitHub <#13 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAGY7CFHV3EIXRIHJP6EIUDVS4IJ7ANCNFSM523JGZXQ . You are receiving this because you were mentioned.Message ID: @.>

BIMSBbioinfo / ikarus

incorrectly labels normal epithelial cells as tumor cells #13